What is Deep Dream?

Does the term ‘Deep Dream‘ spark your imagination? Doesn’t it bring to mind an artistic feeling, like embarking on a journey through a land of fantasy?
Today, we’re diving into the captivating world of Deep Dream. It’s not just a charming phrase; once you start using it, it’s hard to stop exploring its possibilities.

Deep Dream Generator

Developed by Google in 2015, the Deep Dream is a computer vision program that’s more of an AI artist. It uses neural networks to analyze and transform images, rumored to be inspired by the movie ‘Inception’. This program utilizes artificial neural networks inspired by the human brain, reinterpreting existing images into something entirely new.

Deep Dream operates on ‘Deep Learning’ technology, specifically using Convolutional Neural Networks (CNN) to emphasize and exaggerate visual features, resulting in unique interpretations.

The outcome, often abstract and dream-like, led to the naming of this process as ‘Deep Dream’. This algorithm transforms images into dream-like versions, where ordinary images are reinterpreted into exaggerated, surrealistic forms.

Understanding this algorithm requires knowledge of deep learning neural networks. Unlike typical training, which minimizes loss using gradient descent, Deep Dream uniquely trains by maximizing loss through gradient ascent, modifying pixel values directly.

Deep Dream comprehends various features of an image through different layers of the neural network. Simpler patterns like lines and edges in lower layers, and complex visual features like faces and expressions in higher layers. Maximizing image loss can lead to chaos, so training is usually focused on selected layers, influencing the final outcome.

The process involves using identified patterns to transform the image, often resulting in repetitive, distorted, and fantastical effects. As a result, Deep Dream’s creations often morph into surreal, unrealistic images, distinct from their originals. This exploration into AI’s interpretation of visuals opens new avenues in digital art, showcasing how technology can reinvent our perception of reality.

1. How to use?

When you visit the Deep Dream Generator website at https://deepdreamgenerator.com/, you can sign up or log in from the top right corner. I easily created an account by linking my Google account.

Upon first signing up at the Deep Dream Generator website, you receive 100 ‘lightnings’ (visible at the top right corner).
Typically, creating one image consumes about 4 to 40 lightnings, with higher quality images requiring more lightnings.

Then, I’ll try making an image.

1. Write the image you want to create in the [Text Prompt] section.

I entered, “A portrait of a beautiful female knight in silver armor with intricate golden details.” Then, I chose the [Ai Model] as Artistic, changed to high quality, left the rest as default, and clicked the [Generate] icon to create the image.

Wow, the detail quality is better than I expected. Now, I’ll try generating different images with some additional options.

• [1st Image] Set only to High Quality
• [2nd Image] Set only to Add Modifier
• [3rd Image] Set both Add Modifier and High Quality
• [4th Image] Changed Ai Models

2. If there are elements in the image you don’t want, click [Negative].

A new box will appear below [Text Prompt], where you can write them down. To see diversity, I’ll proceed without selecting this option.

There are other detailed options available as well.

3. Click on the [Add Modifiers] icon to set more specific details.

4. For the [Ai Model], I selected Artistic as before, but did not choose high quality.

• *[Ai Model]: Choose the style of art.
• *[Aspect Ratio]: The proportion of width to height.
• *[Quality]: Checking the high-quality box [v] increases lightning usage (by +6 or more).
• *[Face Enhance]: Emphasizes facial features.
• *[Upscale & Enhance]: Corrects noise, increases resolution and clarity, with each step up increasing lightning usage.

I personally think that selecting High Quality is essential. I’ve tried generating images with various other option settings as well.

It seems there’s a clear difference in vivid colors, focusing, and lighting with High Quality. Since the lightning usage for the High Quality option is around 5-10, I’m convinced that I should always choose High Quality in the future.

It’s natural that the image generated varies depending on the option settings. While the Deep Dream Generator theoretically has the unique capability to create surreal and fantastical images, the actual images produced, especially those based on people, seem to possess realistic characteristics influenced by existing data. If one desires to create more unreal, surrealistic images, training the system with examples of surrealist art could yield more desired results.

2. Output

#1. 

• Model: Artistic2
• Prompt: Strange little creature
• Modifiers: bokeh, dof, realistic photo

#2. 

• Model: Artistic2
• Prompt: A mystical wood dwelling inhabited by strange creatures.
• Modifiers: highly detailed, science fiction art, soft lighting

Today, we explored the new AI Art Generator, and isn’t it just delightful? It’s simple, yet the ability to create your own beautiful artwork through a bit of customization always excites me. I encourage you all to try Deep Dream right now and create your own amazing results!

Additionally, if you’re interested in another AI Art Generator, Leonardo AI, feel free to explore more through the link provided below~!

–> Go to Post (What is Leonardo AI and How to use)